On the Adversarial Robustness of Causal Algorithmic Recourse
This addresses the reliability of recourse recommendations for individuals affected by automated decision-making systems, representing an incremental improvement by focusing on adversarial robustness.
The paper tackles the problem of algorithmic recourse recommendations being vulnerable to small uncertainties in individual features, showing that minimally costly recourse methods are not robust. It presents methods for generating adversarially robust recourse for linear and differentiable classifiers and demonstrates that regularizing classifiers to be locally linear and rely on actionable features enables robust recourse.
Algorithmic recourse seeks to provide actionable recommendations for individuals to overcome unfavorable classification outcomes from automated decision-making systems. Recourse recommendations should ideally be robust to reasonably small uncertainty in the features of the individual seeking recourse. In this work, we formulate the adversarially robust recourse problem and show that recourse methods that offer minimally costly recourse fail to be robust. We then present methods for generating adversarially robust recourse for linear and for differentiable classifiers. Finally, we show that regularizing the decision-making classifier to behave locally linearly and to rely more strongly on actionable features facilitates the existence of adversarially robust recourse.